Computer Security http://security.di.unimi.it/sicurezza1819/ - - PowerPoint PPT Presentation

Chapter 3: 1

Computer Security

http://security.di.unimi.it/sicurezza1819/

Chapter 17: 2

Chapter 17: Network Security

Chapter 17: 3

Agenda

▪ Net adversary ▪ TCP attacks ▪ DNS attacks ▪ Firewalls ▪ Intrusion detection ▪ Honeypots

Chapter 17: 4

TCP Session Hijacking

▪ Predict challenge to send messages that appear to

come from a trusted host.

First warning 1984

SYN x SYN ACK x+1, y ACK y+1, x+1 TCP handshake SYN x (spoofed sender) SYN ACK x+1, y ACK y+1, x+1 TCP session hijacking

Chapter 17: 5

TCP SYN Flooding Attacks

▪ Exhaust responder’s resources by creating half-open

TCP connection requests.

y SYN x SYN ACK x+1,y ACK y+1, x+1

TCP handshake

y SYN x SYN ACK x+1,y

SYN flooding attack

y ’ SYN x’

SYN ACK x’+1,y’

. . .

Chapter 17: 6

SYN Cookie Defense

Chapter 17: 7

Domain Name System (DNS)

▪ Essential infrastructure for the Internet. ▪ Maps host names to IP addresses (and vice versa). ▪ Originally designed for a friendly environment; hence

• nly basic authentication mechanisms.

▪ Historic note: DNS created in the 1980s (e.g., RFC

819, August 1982); strong political obstacles to globally deployable cryptographic protection.

▪ Some serious attacks reported in recent years. ▪ We will look at those attacks and at available

countermeasures.

Chapter 17: 8

Domain Name System – DNS

▪ Distributed directory service for domain names (host

names) used for:

➢ look up IP address for host name, host name for IP address. ➢ anti-spam: Sender Policy Framework uses DNS records. ➢ basis for same origin policies applied by web browsers.

▪ Various types of resource records. ▪ Host names and IP addresses collected in zones

managed by authoritative name servers.

▪ Protocols such as BIND, MSDNS, PowerDNS,

DJBDNS, for resolving host names to IP addresses.

▪ We will explain issues at a general, simplified level.

Chapter 17: 9

DNS Infrastructure

▪ 13 root servers; all name servers configured with the

IP addresses of these root servers.

▪ Global Top Level Domain (GTLD) servers for top

level domains: .com, .net, .cn, …

➢ There can be more than one GTLD server per TLD. ➢ Root servers know about GTLD servers.

▪ Authoritative name servers provide mapping between

host names and IP addresses for their zone.

➢ GTLD servers know authoritative name servers in their TLD.

▪ Recursive name servers pass client requests on to

• ther name servers and cache answers received.

Chapter 17: 10

IP Address Lookup – Simplified

▪ Client sends request to its local recursive name server asking to

resolve a host name (target).

▪ Recursive name server refers request to one of the root servers. ▪ Root server returns list of GTLD servers for the target’s TLD; also

sends glue records that give the IP addresses of those servers.

▪ Recursive name server refers request to one of the GTLD servers. ▪ GTLD server returns list of authoritative name server for the

target’s domain, together with their IP addresses (glue records).

▪ Local recursive name server refers the request to one of the

authoritative name servers.

▪ Authoritative mail server provides authoritative answer with IP

address to local name server.

▪ Local recursive name server sends answer to client.

Chapter 17: 11

Name resolution

www.foo.com: 1.2.3.4 www.foo.com ? QID = 2701 www.foo.com ? list of GTLD servers for .com with IP addresses; QID = 2701 list of authoritative name servers for foo.com with IP addresses; QID = 2702 www.foo.com: 1.2.3.4; QID = 2703

recursive ns client: www.foo.com authoritative ns GTLD server Root name server

www.foo.com? QID = 2702 www.foo.com? QID = 2703

Chapter 17: 12

Cache & Time-to-live

▪ Simplified description left out an important aspect. ▪ Performance optimisation: when name server

receives an answer, it stores answer in its cache.

▪ When receiving a request, name server first checks

whether answer is already in its cache; if this is the case, the cached answer is given.

▪ Answer remains in cache until it expires; time-to-live

(TTL) of answer is set by sender.

▪ Design question: reasons for setting TTL by sender,

reasons for setting TTL by receiver?

▪ Long TTL = high security, low TTL = low security?

Chapter 17: 13

Light-weight Authentication

▪ Messages on Internet cannot be intercepted; attacker

can only read messages forwarded to her.

▪ Anybody can pretend to be an authoritative name

server for any zone.

▪ How does a recursive name server know that it has

received a reply from an authoritative name server?

▪ Recursive name server includes a 16-bit query ID

(QID) in its requests.

▪ Responding name server copies QID into its answer;

applies also to answer from authoritative name server.

Chapter 17: 14

Authentication – Security?

▪ If query is not passed by mistake to the attacker, her

chance of generate faking a answer is 2-16.

▪ If

➢ root servers entries at the local name server are correct, ➢ routing tables in the root servers are correct, ➢ routing tables in the GTLD servers are correct, ➢ cache entries at recursive name server are correct,

the attacker will not see original query ID.

▪ Security relies on the assumption that routing from

local recursive name server to authoritative name server is correct.

▪ Attack method: guess QID to subvert cache entries.

Chapter 17: 15

Compromising Authentication

▪

If routing to and from root servers and GTLD servers cannot be compromised, the attacker can only try to improve her chances of guessing a query ID.

▪

Some (earlier) versions of BIND used a counter to generate the QID.

▪

Cache poisoning attack:

1.

Ask recursive name server to resolve host name in attacker’s domain.

2.

Request to attacker’s name server contains current QID.

3.

Ask recursive name server to resolve host name you want to take over; send answer that includes next QID and maps host name to your chosen IP address.

Chapter 17: 16

Predictable Challenges

▪ Lesson: If you want to perform authentication without

cryptography, do not use predictable challenges.

▪ More ways of improving the attack’s chances:

➢ To account for other queries to the recursive name server

concurrent to the attack, send answers with QIDs from a small window.

➢ To increase the chance that fake answer arrives before

authoritative answer, slow down authoritative name server with a DoS attack.

➢ To prevent that a new query for the host name restores the

correct binding, set a long time to live.

Chapter 17: 17

Bailiwick Checking

▪ Performance optimization: name servers send

additional resource records to recursive name server, just in case they might come useful.

▪ Might save round trips during future name resolution. ▪ Works fine if all name servers are well behaved. ▪ Do not trust your inputs: malicious name server might

provide resource records for other domains, e.g. with IP addresses of its choice.

▪ Bailiwick checking: additional resource records not

coming from the queried domain, i.e. records “out of bailiwick”, not accepted by recursive name server.

Chapter 17: 18

DNS Attack – Next Try

▪ Attacker in a race with authoritative name server. ▪ If authoritative answer comes first, the attacker’s next

attempt has to wait until TTL expires.

▪ Attacker does not ask for www.foo.com but for a host

random.foo.com that is not in recursive name server’s cache; triggers a new name resolution request.

➢ Defeats TTL as a measure to slow down attacker; ➢ TTL not intended as a security mechanism!

▪ Authoritative name server for foo.com unlikely to have

entry for random.foo.com.

▪ NXDOMAIN answer indicating that host doesn’t exist.

Chapter 17: 19

Dan Kaminsky’s Attack (2008)

▪ Attacker sends requests for random.foo.com to

recursive name server.

▪ Recursive name server refers request to authoritative

name server for foo.com.

▪ Attacker sends answers for random.foo.com with

guessed QIDs and additional resource record for www.foo.com (in bailiwick).

▪ If guessed QID is correct and attacker’s answer wins

race with NXDOMAIN, entry www.foo.com is cached with a TTL set by attacker.

▪ Recursive name server will now direct all queries for

www.foo.com to attacker’s IP address.

Chapter 17: 20

Dan Kaminsky’s Attack

recursive name server attacker authoritative name server requests for random.foo.com requests for random.foo.com with query ID answers for random.foo.com with guessed QID and RR for www.foo.com; attacker wins race if correct guess arrives before NXDOMAIN. next try, new host NXDOMAIN wins

Chapter 17: 21

Severity of Attack

▪ Very serious attack: attacker becomes name server

for domains of her choice.

▪ Attack increases chance of guessing a QID correctly

by trying many random host names.

▪ Reportedly success within 10 seconds. ▪ Many ways for triggering name resolution at recursive

name server.

▪ Alternative attack strategy: send many faked name

server redirects for www.foo.com with guessed QID (version in Kaminsky’s black hat talk).

Chapter 17: 22

Countermeasures

▪ Increase search space for attacker: run queries on

random ports.

➢ Attacker now must guess QID & port number.

▪ Restrict access to local recursive name server: split

name server (split-split name server).

▪ Trust levels for resource records: access control to

prevent unauthorized overwriting of authoritative data.

▪ DNSSec: cryptographic authentication using digital

signatures; give up on QID as a security feature.

▪ Name server does not reply to malformed queries??

➢ Actually helps the attacker.

Chapter 17: 23

DNS Rebinding Attacks

Chapter 17: 24

DNS Rebinding

▪ Same origin policy: script in a web page can only

connect back to the server it was downloaded from.

▪ To make a connection, the client’s browser needs the

IP address of the server.

▪ Authoritative DNS server resolves ‘abstract’ DNS

names in its domain to ‘concrete’ IP addresses.

▪ The client’s browser ‘trusts’ the DNS server when

enforcing the same origin policy.

▪ Trust is Bad for Security!

Chapter 17: 25

DNS Rebinding Attack

▪ “Abuse trust”: Attacker creates attacker.org domain;

binds this name to two IP addresses, to its own and to the target’s address.

▪ Client downloads applet from attacker.org; script

connects to target; permitted by same origin policy.

▪ Defence: Same origin policy with IP address.

➢ D. Dean, E.W. Felten, D.S. Wallach: Java security: from

HotJava to Netscape and beyond, 1996 IEEE Symposium on Security & Privacy.

Chapter 17: 26

DNS Rebinding Attack

▪ Client visits attacker.org; attacker’s DNS server

resolves this name to attacker’s IP address with short time-to-live.

▪ Attack script waits before connecting to attacker.org. ▪ Binding at browser has expired; new request for IP

address of attacker.org, now bound to target address.

▪ Defence: Don’t trust the DNS server on time-to-live;

pin host name to original IP address;

➢ J. Roskind: Attacks against the Netscape browser. in RSA

Conference, April 2001.

➢ Duration of pinning is browser dependent.

Chapter 17: 27

DNS Rebinding Attack

▪ Attacker shuts down its web server after the page has

been loaded.

▪ Malicious script sends delayed request to attacker.org. ▪ Browser’s connection attempt fails and pin is dropped. ▪ Browser performs a new DNS lookup and is now given

the target’s IP address.

▪ General security issue: error handling procedures

written without proper consideration of their security implications.

Chapter 17: 28

DNS Rebinding Attack

▪ More sophisticated authorisation system: Client

browser refers to policy obtained from DNS server when deciding on connection requests.

▪ Defence: don’t ask DNS server for the policy but the

system with the IP address a DNS name is being resolved to.

➢ Related to reverse DNS lookup. ➢ Similar to defences against bombing attacks in network

security.

Chapter 17: 29

Firewalls

Chapter 17: 30

Introduction

▪ Cryptographic mechanisms protect data in transit

(confidentiality, integrity).

▪ Authentication protocols verify the source of data. ▪ We may also control which traffic is allowed to enter

• ur system (ingress filtering) or to leave our system

(egress filtering).

▪ Access control decisions based on information like

addresses, port numbers, ...

Chapter 17: 31

Firewall

▪ Firewall: a network security device controlling traffic

flow between two parts of a network.

▪ Often installed between an entire organisation’s

network and the Internet.

▪ Can also be installed in an intranet to protect

individual departments.

▪ All traffic has to go through the firewall for protection

to be effective.

➢ Dial-in lines, wireless LANs, USB devices!?

Chapter 17: 32

Purpose

▪ Firewalls control network traffic to and from the

protected network.

▪ Can allow or block access to services (both internal

and external).

▪ Can enforce authentication before allowing access to

services.

▪ Can monitor traffic in/out of network.

Chapter 17: 33

Types of Firewalls

▪ Packet filter ▪ Stateful packet filter

Chapter 17: 34

Packet Filter

▪ Inspect headers of IP packets, also TCP and UDP port

numbers.

▪ Rules specify which packets are allowed through the

firewall, and which are dropped.

➢ Actions: bypass, drop, protect (IPsec channel).

▪ Rules may specify source / destination IP addresses,

and source / destination TCP / UDP port numbers.

▪ Rules for traffic in both directions. ▪ Certain common protocols are difficult to support

securely (e.g. FTP).

Chapter 17: 35

Example

▪ TCP/IP packet filtering router.

➢ Router which can throw packets away.

▪ Examines TCP/IP headers of every packet going

through the Firewall, in either direction.

▪ Packets can be allowed or blocked based on:

➢ IP source & destination addresses ➢ TCP / UDP source & destination ports

▪ Implementation on router for high throughput.

Chapter 17: 36

Stateful Packet Filter

▪ Packet filter that understands requests and replies

(e.g. for TCP: SYN, SYN-ACK, ACK).

▪ Rules need only specify packets in one direction

(from client to server – the direction of the first packet in a connection).

▪ Replies and further packets in the connection are

automatically processed.

▪ Supports wider range of protocols than simple packet

filter (eg: FTP, IRC, H323).

Chapter 17: 37

Firewall Policies

▪ Permissive: allow by default, block some.

➢ Easy to make mistakes. ➢ If you forget something you should block, it’s allowed, and

you might not realise for a while.

➢ If somebody finds find a protocol is allowed, they might not

tell you ....

▪ Restrictive: block by default, allow some.

➢ Much more secure. ➢ If you forget something, someone will complain and you can

allow the protocol.

Chapter 17: 38

Firewall Policies – Eexamples

▪ Permissive policies: Allow all traffic, but block ...

➢ Irc ➢ telnet ➢ snmp ➢ …

▪ Restrictive policies: block all traffic, but allow ...

➢ http ➢ Pop3 ➢ Smtp ➢ ssh ➢ …

Chapter 17: 39

Rule Order

▪ A firewall policy is a collection of rules. ▪ Packets can contain several headers (→ IPsec). ▪ When setting a policy, you have to know in which

• rder rules (and headers) are evaluated.

▪ Two main options for ordering rules:

➢ Apply first matching entry in the list of rules. ➢ Apply the entry with the best match for the packet.

Chapter 17: 40

Typical Firewall Ruleset

▪ Allow from internal network to Internet:

➢ HTTP, FTP, HTTPS, SSH, DNS

▪ Allow reply packets ▪ Allow from anywhere to Mail server:

➢ TCP port 25 (SMTP) only

▪ Allow from Mail server to Internet:

➢ SMTP, DNS

▪ Allow from inside to Mail server:

➢ SMTP, POP3

▪ Block everything else

Chapter 17: 41

Firewall Location

▪ Firewall can only filter traffic which goes through it. ▪ Where to put, for example, a mail server? ▪ Requires external access to receive mail from the

Internet.

➢ Should be on the inside of the firewall

▪ Requires internal access to receive mail from the

internal network.

➢ Should be on the outside of the firewall

▪ Solution: “perimeter network” (aka DMZ).

Chapter 17: 42

DMZ

DMZ

inner firewall

• uter

firewall Inter net local network Web server mail server

Chapter 17: 43

Firewalls – Limitations

▪ Firewalls do not protect against insider threats. ▪ Blocking services may create inconveniences for

users.

▪ Network diagnostics may be harder. ▪ Some protocols are hard to support. ▪ Protocol tunnelling: sending data for one protocol

through another protocol circumvents the firewall.

➢ As more and more administrators block almost all ports but

have to leave port 80 open, more and more protocols are tunnelled through http to get through the firewall.

▪ Encrypted traffic cannot be examined and filtered.

Chapter 17: 44

Intrusion Detection Systems

Chapter 17: 45

Reminder: Security Strategies

▪ Prevention: take measures that prevent your assets

from being damaged.

▪ Detection: take measures so that you can detect

when, how, and by whom an asset has been damaged.

▪ Reaction: take measures so that you can recover

your assets or to recover from a damage to your assets.

Chapter 17: 46

Comment

▪ Cryptographic mechanisms and protocols are fielded

to prevent attacks.

▪ Perimeter security devices (e.g. firewalls) mainly

prevent attacks by outsiders.

▪ Although it would be nice to prevent all attacks, in

reality this is rarely possible.

▪ New types of attacks occur: denial-of-service (where

crypto may make the problem worse).

▪ We will now look at ways of detecting network

attacks.

Chapter 17: 47

Intrusion Detection Systems

▪ Passive supervision of network, analogue to intruder

alarms.

➢ Creates more work for personnel. ➢ Provides security personnel with volumes of reports that can

be presented to management …

▪ Two approaches to Intrusion Detection:

➢ Knowledge-based IDS – Misuse detection ➢ Behaviour-based IDS – Anomaly detection

▪ Network based and host based IDS. ▪ Given the (near) real-time nature of IDS alerts, an

IDS can also be used as response tool.

Chapter 17: 48

Knowledge-based IDS

▪ Knowledge-based IDS looks for patterns of network

traffic or activity in log files that indicate suspicious behaviour, using information such as:

➢ known vulnerabilities of particular OS and applications; ➢ known attacks on systems; ➢ given security policy.

▪ Example signatures might include:

➢ number of recent failed login attempts on a sensitive host; ➢ bit patterns in an IP packet indicating a buffer overrun attack; ➢ certain types of TCP SYN packets indicating a SYN flood

DoS attack.

▪ Also known as misuse detection IDS.

Chapter 17: 49

Knowledge-based IDS

▪ Only as good as database of attack signatures:

➢ New vulnerabilities not in the database are constantly being

discovered and exploited;

➢ Vendors need to keep up to date with latest attacks and

issue database updates; customers need to install these;

➢ Large number of vulnerabilities and different exploitation

methods, so effective database difficult to build;

➢ Large database makes IDS slow to use.

▪ All commercial IDS look for attack signatures.

Chapter 17: 50

Behaviour-based IDS

▪ Wouldn’t it be nice to be able to detect new attacks? ▪ Statistical anomaly detection uses statistical

techniques to detect attacks.

▪ First establish base-line behaviour: what is “normal”

for this system?

▪ Then gather new statistical data and measure

deviation from base-line.

▪ If a threshold is exceeded, issue an alarm. ▪ Also known as behaviour-based detection.

Chapter 17: 51

Behaviour-based IDS

▪ Example: monitor number of failed login attempts at a

sensitive host over a period;

➢ if a burst of failures occurs, an attack may be under way; ➢ or maybe the admin just forgot his password?

▪ False positives (false alarm): attack flagged when

none is taking place.

➢ See e.g. Richard Bejtlich: Interpreting Network Traffic: A

Network Intrusion Detector’s Look at Suspicious Events, Proceedings of the 12th Annual Computer Security Incidence Handling Conference, Chicago, 2000.

▪ False negatives: attack missed because it fell within

the bounds of normal behaviour.

Chapter 17: 52

Anomaly Detection

▪ IDS does not need to know about security

vulnerabilities in a particular system:

➢ base-line defines normality; ➢ IDS does not need to know details of the construction of a

buffer overflow packet.

▪ Anomalies are not necessarily attacks; normal and

forbidden behaviour may overlap:

➢ Legitimate users may deviate from baseline, causing false

positives (e.g. user goes on holiday, works late in the office, forgets password, or starts to use new application).

➢ If base-line is adjusted dynamically and automatically, a

patient attacker may be able to gradually shift the base-line

• ver time so that his attack does not generate an alarm.

Chapter 17: 53

IDS Architecture

▪ Distributed set of sensors – either located on hosts or

• n network – to gather data.

▪ Centralised console to manage sensor network,

analyze data (→ data mining), report and react.

▪ Ideally:

➢ Protected communications between sensors and console; ➢ Protected storage for signature database/logs; ➢ Secure console configuration; ➢ Secured signature updates from vendor; ➢ Otherwise, the IDS itself can be attacked and manipulated;

IDS vulnerabilities have been exploited.

Chapter 17: 54

HIDS & NIDS

▪ Network-based IDS (NIDS) looks for attack

signatures in network traffic.

▪ Host-based IDS (HIDS) looks for attack signatures in

log files of hosts.

▪ Trend towards host-based IDSs. ▪ Attacks a NIDS can detect but a HIDS cannot:

➢ SYN flood, Land, Smurf,Teardrop, BackOrifice,…

▪ And vice-versa:

➢ Trojan login script, walk up to unattended keyboard,

encrypted traffic,…

▪ For more reliable detection, combine both IDS types.

Chapter 17: 55

Network-based IDS

▪ Uses network packets as data source. ▪ Typically a network adapter running in promiscuous

mode.

▪ Monitors and analyzes all traffic in real-time. ▪ Attack recognition module uses three common

techniques to recognize attack signatures:

➢ Pattern, expression or bytecode matching; ➢ Frequency or threshold crossing (e.g. detect port scanning

activity);

➢ Correlation of lesser events (in reality, not much of this in

commercial systems).

Chapter 17: 56

Placement of NIDS

Internet

Firewall

Mail server Web server Console

sensor sensor sensor protected network perimeter network

Chapter 17: 57

Host-based IDS

▪ Typically monitors system, event, and security logs

• n Windows and syslog in Unix environments.

➢ E.g., observe sequences of system calls to check whether a

change from user to supervisor mode had been effected properly through a command like su.

▪ Verify checksums of key system files & executables

at regular intervals for unexpected changes.

▪ Some products use regular expressions to refine

attack signatures;

➢ E.g., passwd program executed AND .rhosts file changed.

▪ Some products listen to port activity and alert when

specific ports are accessed – limited NIDS capability.

Chapter 17: 58

Placement of HIDS

Internet

Firewall

Mail server Web server Console

sensor sensor sensor internal network perimeter network

Chapter 17: 59

Honeypots

▪ How to detect zero-day exploits? There is no attack

signature yet.

▪ How to “collect” new attacks for the knowledge base? ▪ Put systems online that mimic production systems but

do not contain “real” data; anything observed on these systems is an attack.

▪ Honeypot: “… a resource whose value is being

attacked or compromised”

➢ Laurence Spitzner, “The value of honeypots”, SecurityFocus,

October 2001

▪ Honeypot: technology to track, learn and gather

evidence of hacker activities.

Chapter 17: 60

Honeypot Types

▪ Level of Involvement:

➢ Low interaction: port listeners ➢ Mid interaction: fake daemons ➢ High interaction: real services

▪ Quality of information acquired increases with level of

interaction.

▪ ‘Intelligent’ attackers will avoid obvious honeypots;

tools for detecting honeypots exist.

▪ Risk that honeypot can be used as staging post in an

attack increases with level of interaction.

Chapter 17: 61

Honeynet

▪ Network of honeypots. ▪ Supplemented by firewalls and intrusion detection

systems – Honeywall.

▪ Advantages:

➢ “More realistic” environment ➢ Improved possibilities to collect data

Chapter 17: 62

Summary

▪ Apply prevention, detection and reaction in

combination.

▪ IDS useful second line of defence (in addition to

firewalls, cryptographic protocols, etc.).

▪ IDS deployment, customisation and management is

generally not straightforward.

▪ Anomalies are not necessarily attacks.